Mining Matrix Pattern from Mobile Users - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 250

in the y axis. The content of the matrix is generated through the random number generator function in
Matlab. Performance evaluation is charted using Matlab, Microsoft Excel and SPSS.
Performance of Single Layer Matrix
Figure 16 shows the overall comparison among the different configuration in terms of Y coordinates in
single layer matrix. It can be seen that the gradients are different, and they are increasing at a different
rate. A fine tuning of the total number of X and Y coordinates is required depending on the physical
environment, such as the geographical largeness of the area concerned. Generally, the higher the reso-
lution, the better the identification process and the coordinates can further be rounded back to specific
reference point of physical subjects for mobile user data mining. The second chart shows the graphical
comparison among the net location identifiers for different configurations of single layer matrix at X =
30. The dataset named Equal refers to X = Y , Double refers to Y = 2 X , and Triple refers to Y = 3 X .
It appears that there exists a trade-off between the greater resolution (greater net location identifiers)
and the ability to accurately reference the physical location points. As the resolution increases, that is,
the increase of the number of either X or Y coordinates, it will return higher net location identifiers. The
resulting increase of the net location identifiers increases the ambiguity of physical location reference,
as a very fine physical location point is specified in the data mining process, while the actual subjects
may occupy many of the reference points.
There are two options, that is, either use multiple net location points for the data mining system,
while accepting the risk of inaccuracies due to variability of physical location reference point, or use a
single fixed physical subject reference point, such as a reference point that is centre, and nearest to all
border reference points of a physical subject. The second method is more desirable, and can be achieved
by first rounding the physical location reference points of physical visit transactions from mobile nodes.
For instance, a mobile node that pass by a static node will have the coordinates rounded to the average
fixed reference point of the static node, so that it provides a standardized, and consistent reference point
for measuring relative physical relationship strengths.
The performance evaluation also shows the increasing number of possible coordinates that can be
used for identifying a particular physical location in the matrix. However, the greater this number, the
better for the data mining system, as the integrity of the knowledge depends on whether the system
can deliver high resolution matrix. The higher the resolution, the more fine the details that the matrix
can support. Due to the fact that reference to these identifiers are done by using integer variables, the
consideration is merely a single, double or triple digit figures, which can be considered negligible in
todays terms.
The previous performance evaluation shows the relationships among the number of XY coordinates
and the resolution of the matrix pattern. It has been shown that the net physical identifier increases at
a gradual rate as the Y coordinate increases. The following section will be focused on the comparison
of capabilities among matrix pattern and other previous mobile user data mining methods, such as
frequency pattern, group pattern, and parallel pattern.
The comparison among matrix pattern and other possible methods involves the analysis on how the
problem is solved. As this is a comparison between different strategies of solving a problem, the ef-
ficiency of a strategy must first be analysed. For matrix pattern strategy, it has been described. If group
pattern strategy is used, group pattern is unable solve such a problem, as it involves finding knowledge
by forming a group of mobile users. For frequency pattern, it also cannot solve this problem, as frequency
pattern is frequency of communication oriented.